Cyfip1 haploinsufficiency is associated with white matter changes, myelin thinning, reduction of mature oligodendrocytes and behavioural inflexibility

The biological basis of the increased risk for psychiatric disorders due to the pathogenic Previous work has shown disturbances in white matter tracts in human carriers of the deletion at this locus. Using a novel rat model, we have recapitulated low dosage of the candidate risk gene CYFIP1 present within the 15q11.2 interval to reveal alterations in white matter microstructure. Using diffusion tensor imaging, we first showed extensive white matter changes in Cyfip1 mutants that were most pronounced in the corpus callosum and external capsule. Transmission electron microscopy showed that these changes were associated with thinning of the myelin sheath in the corpus callosum. Myelin thinning was independent of changes in axon number or diameter but was associated with a reduction in the number of mature oligodendrocytes. Finally, we demonstrated functional effects on cognitive phenotypes sensitive to both disruptions in myelin and callosal circuitry. significant decreased myelin thickness in Cyfip1 +/- rats. f mean g-ratios calculated for small (n=1510 WT and 1276 Cyfip1 +/- axons; LME: c 2 (1)=4.23*), medium-small (n=2283 WT and 2043 Cyfip1 +/- axons; LME: c 2 (1)=4.44,*), medium-large (n=2551 WT and 1993 Cyfip1 +/-axons; LME: c 2 (1)=7.14,**), and large (n=804 WT and 667 Cyfip1 +/- axons; LME: c 2 (1)=13.92,***) myelinated axons, showing significant increases in g-ratio in all different axon diameter ranges, and more significant in larger axons. g Scatter plot of myelin thickness values across all axon diameters WT (n=7148 axons) and Cyfip1 +/- (n=5980 axons). Differences between axon diameter, g-ratio and myelin thickness measures were assessed using linear mixed models adjusted for individual variability. Data are mean ± SEM; *<0.05,


Abstract
The biological basis of the increased risk for psychiatric disorders due to the pathogenic 15q11.2 copy number deletion is unknown. Previous work has shown disturbances in white matter tracts in human carriers of the deletion at this locus. Using a novel rat model, we have recapitulated low dosage of the candidate risk gene CYFIP1 present within the 15q11.2 interval to reveal alterations in white matter microstructure. Using diffusion tensor imaging, we first showed extensive white matter changes in Cyfip1 mutants that were most pronounced in the corpus callosum and external capsule. Transmission electron microscopy showed that these changes were associated with thinning of the myelin sheath in the corpus callosum.
Myelin thinning was independent of changes in axon number or diameter but was associated with a reduction in the number of mature oligodendrocytes. Finally, we demonstrated functional effects on cognitive phenotypes sensitive to both disruptions in myelin and callosal circuitry.
. CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; Low gene dosage of the cytoplasmic FMRP interacting protein 1 (CYFIP1) gene is a candidate risk factor for psychopathology by virtue of its involvement in the pathogenic 15q11.2 BP1-BP2 copy number variant (CNV). Heterozygous deletion of this genomic interval leads to a two-to four fold increase in risk for intellectual disability and psychiatric problems, including schizophrenia, autism, as well as a significant increase in the risk for epilepsy 1,2 . The deletion contains four genes: non-imprinted in Prader-Willi/Angelman syndrome 1 gene (NIPA1), non-imprinted in Prader-Willi/Angelman syndrome 2 gene (NIPA2), CYFIP1, and tubulin gamma complex associated protein 5 gene (TUBGCP5) 3 .
Whilst all these genes are expressed in the brain and may be of potential relevance to psychopathology, CYFIP1 haploinsufficiency is considered to be a likely significant contributor to the 15q11.2 BP1-BP2 psychiatric phenotype due to its known involvement in a number of key brain plasticity-related functions. These include alterations in dendritic spine morphology and branching, mediated by interactions in two distinct complexes: the WAVE regulatory complex to modulate ARP2/3 dependent actin cytoskeleton dynamics, and CYFIP1-eIF4E complex to suppress protein translation at the synapse through interactions with fragile X mental retardation 1 protein (FMRP), the gene product of FMR1 4 . Mutations in FMR1 are causative for fragile X syndrome, a condition associated with intellectual disability and a range of psychiatric symptoms 5 .
Changes in white matter microstructure have been reported consistently in major psychiatric disorders including schizophrenia, autism and intellectual disability 6 . Moreover, using diffusion tensor imaging (DTI) methods, we found extensive white matter changes in 15q11.2 BP1-BP2 CNV carriers, specifically widespread increases in fractional anisotropy (FA) in deletion carriers (in press) 7 . Some of the biggest changes we observed were in the posterior limb of the internal capsule and corpus callosum. Prominent effects in the corpus . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; callosum are consistent with previous findings by others of increased corpus callosum volume in 15q11.2 BP1-BP2 deletion subjects 8  Our focus on white matter microstructure was also guided by evidence that CYFIP1 is an actin regulator, and thus likely to affect white matter via the requirement of precise regulation of the actin cytoskeleton for normal cellular development, morphology and migration. Hence, CYFIP1 haploinsufficiency has the potential to disrupt axonal organisation via both effects on axonal guidance 9 and the myelin component of white matter tracts 10,11 . Myelin is produced by mature oligodendrocytes and several studies have linked actin regulators to oligodendrocyte-myelin dynamics. The Wiskott-Aldrich Syndrome protein family member 1 (WAVE1) and the integrin-linked kinase (ILK) regulate oligodendrocyte differentiation and axon ensheathment 12,13 , while the Arp2/3 complex, a key actin nucleator, is required for initiation of myelination 11 , and Rho GTPases Cdc42 and Rac1 regulate myelin sheath formation 14 .
We therefore hypothesised there would be white matter abnormalities in the Cyfip1 +/rat line possibly linked to underlying changes in axonal architecture including myelin. We also . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; anticipated functional effects on brain and behaviour on the basis that axon-myelin perturbations can have marked effects on brain network activity caused by disruptions in the temporal coherence of action potential integration across different brain regions 15,16 .
Synchronization of synaptic signals is crucial in learning, and a previous study in shiverer (deletion mutant of myelin basic protein (MBP)) and mld (allelic mutant to shiverer with lowered MBP expression) mice 15 showed that deficits in myelination had a specific effect on cognitive flexibility in a reversal learning task. In the present work, we looked for evidence of maladaptive brain function in the Cyfip1 +/rats using behavioural tasks that assayed cognitive flexibility.
Our findings from in vivo DTI confirmed that Cyfip1 haploinsufficiency is associated with white matter microstructure changes, most prominently in the corpus callosum and external capsule. Consistent with our conjecture that Cyfip1 haploinsufficiency may influence underlying axon-oligodendrocyte-myelin dynamics, transmission electron microscopy demonstrated that the white matter changes in the corpus callosum were associated with a thinning of the myelin sheath. The effects on myelin were independent of changes in axon number or diameter but were associated with a reduction in the population of mature oligodendrocytes, the specific cell-type that produces myelin. Additionally, we demonstrated changes in cognitive phenotypes required for flexible behavioural responses that were consistent with effects on callosal circuitry and representative of core deficits that are prominent across schizophrenia, autism and fragile X syndrome [17][18][19][20][21][22][23] .
. CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

Cyfip1 haploinsufficiency disrupts white matter microstructure
Full details of the creation of the Cyfip1 +/rat model are in the Supplementary Methods. CRISPR/Cas9 targeting led to a 4bp out of frame heterozygous deletion in exon 7 of the Cyfip1 gene at location Chromosome 1: 36974-36977 and a resulting bioinformatics prediction of an early stop codon in exon 8 which was verified functionally using qPCR and Western Blot to measure reductions in mRNA and protein respectively.
To investigate white matter microstructure in the Cyfip1 +/rat brain a cohort of 24 male rats (wild-type (WT) n=12, Cyfip1 +/-n=12) were anaesthetised with isoflurane in oxygen at 4% and maintained at 1%, and DTI data were collected using a 9.4T MRI scanner, utilising 60 noncollinear gradient directions with a single b-value shell at 1000 s/m 2 . Group comparisons were carried out using Tract-Based Spatial Statistics (TBSS) 24 available in FMRIB Software Library (FSL), with a randomize function allowing voxel-wise nonparametric permutation analysis of the DTI maps projected onto a whole brain white matter skeleton (Supplementary Figure 1). The randomize function was used with the threshold-free cluster enhancement (TFCE) 25 , generating cluster-size statistics based on 1000 random permutations. Figure 1 shows the regions where significant differences in white matter microstructure were found after correction for multiple comparisons. Figure 1a shows the pattern of changes using a highly conservative family-wise error (FWE) correction. This approach showed consistent reductions in FA in the corpus callosum, in the external and internal capsule, and in parts of the fimbria/fornix in Cyfip1 +/rats, with no differences in axial diffusivity (AD), radial diffusivity (RD), and mean diffusivity (MD). We complemented the highly conservative . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. correction for multiple comparisons based on the Benjamini-Hochberg procedure 26 used previously by ourselves 27 and others 28 in rodent imaging data. This analysis, shown in Figure   1b, revealed additional white matter changes including increases in FA in regions of the fornix and fimbria suggesting that Cyfip1 haploinsufficiency may have differential effects in different brain regions. Figure 1b also shows changes in other DTI metrics, after FDR correction, illustrating mostly decreases in AD and increases in RD, and MD. These effects were complementary in terms of (a) being localised in the corpus callosum and external and internal capsule and (b) being consistent with the overall predominant effects of Cyfip1 haploinsufficiency in reducing FA.
. CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
All the parametric maps were generated at a significance level of p<0.05. c Scale bars indicating the direction of the changes in both a) and b), where relative decreases in Cyfip1 +/rats are represented by a gradient of blue (less significant) to green (more significant), and CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; relative increases in Cyfip1 +/are represented by a gradient of red (less significant) to yellow (more significant). We next manually generated binary masks of regions of interest (corpus callosum, internal capsule, external capsule and fimbria/fornix), guided by the results from FWE correction using FSL (Supplementary Figure 2), and assessed mean FA, AD, RD and MD in these white matter tracts. As can be seen in Table 1, analysing the DTI data in this way (which averaged differences between WT and Cyfip1 +/rats within a discrete fibre tract, as opposed to the voxel-by-voxel analysis which detected clusters of voxel differences in white matter tracts across the whole brain) showed that the most significant differences were reductions in FA in the corpus callosum (t=2.3, df=20.75, p<0.05) and external capsule (t=2.4, df=22, p<0.05) in the Cyfip1 +/rats compared to WT, as assessed with a two-tailed unpaired t-test. These data were consistent with the previous voxel-by-voxel analysis and provided the additional finding that the most extensive white matter changes in the Cyfip1 +/rats occurred in these structures. . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

Cyfip1 haploinsufficiency is associated with myelin abnormalities in corpus callosum
We next investigated the cellular nature of the Cyfip1 associated DTI changes. DTI measures can be affected by several factors and previous studies have linked decreases in FA in white matter tracts with less myelin, lower axonal density, axonal damage, or changes in axonal organisation 29,30 . To assess cellular changes, we carried out an ultra-structural analysis, blind to genotype, using transmission electron microscopy and focusing on the corpus callosum, given the DTI data indicating the sensitivity of this structure to Cyfip1 haploinsufficiency.
The experiment used a new cohort of rats (WT n=5, Cyfip1 +/-n=4). In order to obtain a representative sample, we sampled 15 regions across the anterior-posterior extent of the corpus callosum encompassing the genu, body and splenium, from sagittal brain sections (representative electron microscopy micrographs in Figure 2a). We measured the number of myelinated and unmyelinated axons, the inner diameter and the outer diameter (including the myelin sheath) of each axon, and then calculated the myelin thickness and the g-ratio (myelin thickness relative to axon diameter, where smaller g-ratios indicate greater myelin thickness) of each myelinated axon (see measures taken in Figure 2b).
We used linear mixed effects (LME) models to analyse the effect of genotype on axon diameter, g-ratio and myelin thickness, considering variation across animals, whereas a twotailed unpaired t-test was used to compare the number of axons between groups. In this analysis, the myelin thickness was log-transformed since the data followed a log-normal distribution, whereas the other measures followed a normal distribution. No genotype CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; calibre in the corpus callosum of the Cyfip1 +/rats. The analyses did not show a significant increase in g-ratio when comparing all axons in each group (LME: c 2 (1)=2.03, p=0.15), however it revealed a significant reduction in myelin thickness in the Cyfip1 +/rats (LME: c 2 (1)=14.63, p<0.001), both shown in Figure 2e. The fact that we did not see a significant difference in g-ratio could have resulted from variability in the average of axon diameters within animals in the same group, which is related to g-ratio (Supplementary Figure 3). Furthermore, we needed to consider that the extent of myelination can be related to axon diameter 31 , and whether the effects on the g-ratio were specific to certain sizes of axons.
Analysing the g-ratio of axons within specific diameter ranges revealed a significant increased g-ratio in each interval in the Cyfip1 +/rats (Figure 2f), that was more significant in larger axons. These analyses indicate decreased myelin thickness in the corpus callosum of the Cyfip1 +/rats that is more pronounced in larger axons.
. CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.  . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

Cyfip1 haploinsufficiency leads to reduced numbers of mature oligodendrocytes in corpus callosum/external capsule
Myelin is produced by mature oligodendrocytes, so we next tested whether Cyfip1 haploinsufficiency influenced the number and/or the maturation of oligodendrocytes using antibodies to the specific molecular markers Olig2 and Cc1. This experiment used rats taken randomly from the same group of rats providing the DTI data shown in Figure 1 (WT n=7 WT and Cyfip1 +/-n=7). The analysis focused on the corpus callosum and external capsule and at least four random sections were taken for quantification in each rat from coronal sections. Sections were stained for Olig2 and Cc1 proteins. Cells stained for Olig2 alone represented all the oligodendrocyte lineages from early progenitors to mature cells, whereas cells double-stained for Olig2 and Cc1 proteins revealed specifically the mature oligodendrocyte (myelin-producing) population. In the Cyfip1 +/rats, this analysis showed a . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; significant reduction in both the number of oligodendrocyte lineage cells (Figure 3a; t=2.18, df=11.94, p<0.05) and mature oligodendrocytes (t=2.48, df=11.99, p<0.05) in comparison with WT. We also found a marginally significant reduction in the level of myelin basic protein (MBP) (Figure 3b, two-tailed unpaired t-test, t=2.16, df=11.96, p=0.052) in the corpus callosum/external capsule of the Cyfip1 +/rats. These data indicated that Cyfip1 haploinsufficiency resulted in reduced numbers of mature oligodendrocytes. We concluded that these data were consistent with effects of low dosage Cyfip1 in reducing myelination that were in turn consistent with the disruptions in white matter microstructure revealed by the DTI analyses. . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

Cyfip1 haploinsufficient rats show deficits in cognitive flexibility.
We next assessed whether the Cyfip1 related imaging and cellular phenotypes were associated with effects on behaviour. Behavioural changes have been observed in rodent models of reduced myelination including shiverer and mld mice with modifications in myelin basic protein 15 . Both mutants showed highly specific effects on behavioural flexibility in a reversal learning task, whereby they had difficulty inhibiting a well learned pre-potent response in order to learn a new response, these effects occurring in the absence of any fundamental learning deficits per se 15 . Furthermore, both human [17][18][19][21][22][23] and animal studies 20,32 have also shown that disruptions to connectivity involving callosal circuitry and circuits involving the internal and external capsules impact on a number of psychological functions, in particular those mediating attention and response control, especially response inhibition.
To assay response inhibition, we utilised a touch screen based appetitive reversal learning task in a separate cohort of experimental rats (WT n=7, Cyfip1 +/-n=10). The reversal learning task first allowed an assessment of basic appetitive learning where rats had to learn that one visual stimulus was associated with reward (the S+) and another stimulus was not (S-), the two stimuli counterbalanced across animals. This was followed by reversal of the . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. contingencies (see Supplementary Figure 4 for flowchart of task design). Successful reversal is demanding on attention and response inhibition, and can be viewed as a type of "go"-"nogo" discrimination task, similar to those that have been used in clinical studies (i.e. the Stroop task, SART, stop signal) to examine response inhibition 33-39 . All rats successfully completed the early stages of pre-training in the reversal learning task where they had to learn to collect food from the magazine and to touch stimuli presented on the touchscreen to earn rewards, Following acquisition of the initial visual discrimination the contingencies were reversed. This manipulation had the expected effect of dramatically reducing correct response as the rats initially persisted in responding to the previously correct but now incorrect (i.e. unrewarded) stimulus leading to 'perseverative' below-chance response. However, and in contrast to the basic learning of the visual discrimination, reversal of the contingencies revealed genotype effects evident in the immediate post-reversal sessions (see Figure 4c). In this part of the task, where inhibition of the previously correct but now incorrect response is the main process controlling behaviour 40,41 , the Cyfip1 +/rats continued to behave according to the previous response strategy showing a significantly greater degree of perseverative, inflexible response, which resulted in them being unable to switch efficiently to the new . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.  In subsequent sessions with the reversed contingencies the rats were gradually able to inhibit the previous response until they reached the criterion of 50% correct, where perseverative below chance responses were no longer apparent. However, again, group differences were apparent in those fewer Cyfip1 +/rats that successfully reached this criterion (Figure 4d, c 2 =9.61, p<0.05). Effectively this sub-group of rats were never able to successfully inhibit the previously learned response despite being given ample opportunity to do so that extended to the end of the experiment (rats that completed this phase of reversal did so in an average of 13 sessions, whereas those that failed to learn the reversed contingencies had an average of 22 sessions before the end of the experiment). Figure 4d also illustrates the high degree of behavioural specificity shown by the Cyfip1 +/rats in the task; insofar as genotype differences were not evident in the relative proportion of those rats that were able to successfully inhibit the previous response strategy, and went on to learn the new contingency to 80% criterion.
Moreover, Cyfip1 +/rats that completed these stages of reversal (reaching 50% and 80% correct) did so in a similar number of sessions to WTs (Figure 4e; see also Supplementary Table 1a for the number of rats completing each stage of the reversal task; the total sessions and trials across the whole task are also shown in Table 1b).
. CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. We further assessed the effects of Cyfip1 haploinsufficiency on flexible responding in another cohort of rats (WT n=21, Cyfip1 +/-n=15) using an associative mismatch task. This task again assayed visual object recognition and discrimination but in addition allowed a determination of novelty detection and habituation, the latter a form of inhibition distinct from that required in reversal learning tasks. Following habituation to the experimental chambers, the rats received presentations of auditory cues that were predictive of different    Table 1 for additional data.

Discussion
We used a CRISPR/Cas9 engineered rat line to model the contribution of Cyfip1 haploinsufficiency to white matter changes observed in carriers of the pathogenic 15q11.2 BP1-BP2 deletion. The Cyfip1 +/rat model allowed us to carry out a DTI analysis with high resolution using identical preprocessing to our 15q11.2 BP1-BP2 human imaging study, and employ rigorous statistics including exploratory voxel-wise assessments permitting comparisons of genotype effects across brain regions. The Cyfip1 +/rat line provided enhanced experimental tractability in terms of direct access to brain tissue, with interpretation of DTI changes at the cellular level, and also allowed relevant behavioural analyses under controlled conditions.
A main finding of the DTI experiments were significant decreases in FA that were most pronounced in the corpus callosum and external capsule. These data were obtained using a highly conservative correction procedure. More widespread changes in white matter were . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; apparent when using a less conservative correction method including increased FA in some areas of the fornix/fimbria. An advantage of using the Cyfip1 +/rat model was that we could make a direct assessment of possible cellular changes underlying the DTI effects. The precise relationship between DTI measures and cellular changes is subject to ongoing debate 30 and whilst DTI can identify white matter changes it cannot definitively distinguish between disruptions to axons and/or myelin 43 . Consequently, at the outset the DTI effects we obtained in the rat model could have been related to changes in axon microstructure or myelin, or both. . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; The effects of Cyfip1 haploinsufficiency on oligodendrocytes and myelination may be related to core deficits in actin physiology, and previous data from Cyfip1 +/mouse models have shown abnormal f-actin dynamics 48 . As noted previously, a main function of CYFIP1 protein is to inhibit ARP2/3 dependent actin cytoskeleton remodelling, via inhibition of WAVE1 protein and other members of the WAVE regulatory complex. It is known that multiple aspects of oligodendrocyte function, including cell proliferation, differentiation and migration are critically reliant on effective cytoskeleton remodelling 10 . Furthermore, alterations of the cytoskeleton are required to produce myelin, being the formation of lamellipodia and lamellipodial 'ruffles', that make initial contact between the oligodendrocyte and axon, required for myelination to occur 49 . It has also been shown in mouse models that manipulating WAVE1 protein directly impacts on the ability to form myelin, and interestingly these effects were localised in the corpus callosum and not present in other areas of the central nervous system 12 . The extent to which the current effects of Cyfip1 haploinsufficiency on oligodendrocyte function are consistent with the effects of WAVE1 manipulation remains to be determined. Effects related to the close interaction between CYFIP1 and FMRP cannot be discounted, especially in view of a degree of overlap between white matter changes in the Cyfip1 +/rat model and a mouse Fmr1 knockout, specifically reduced FA in the corpus callosum 50 . The mouse Fmr1 knockout also revealed evidence of global disruptions in functional connectivity. Assuming a role for Cyfip1-Fmrp interactions, the relevant molecular and cellular mechanisms would presumably involve perturbations to synaptic protein homeostasis, however the precise relationship between such effects on synaptic protein dynamics and the phenotypes we observed in our rat model remain to be identified.
. CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. We assessed the extent to which the imaging and cellular data were associated with functional behavioural effects. Behavioural flexibility has been shown to be sensitive to abnormalities in myelination in mouse models of myelin deficits 15 . Furthermore, in humans, corpus callosum morphology was shown to correlate with response inhibition in a study using a cohort of twin pairs, where one was diagnosed with bipolar disorder and the other was clinically healthy 17 . We therefore focused the behavioural analyses on psychological processes supporting behavioural flexibility. In a first experiment assaying reversal learning, we showed that the Cyfip1 +/rats perseverated, meaning they continued to respond to stimuli that had resulted in a correct response previously but, following reversal, were now associated with an incorrect response. The effects in the Cyfip1 +/rats were highly specific to the reversal element of the task with no effects on initial learning, and indicated the absence of any generalised learning deficits. Instead, the pattern of data suggested a deficit in the ability to flexibly manipulate previous learning to adapt to changing contingencies.
To extend these behavioural data, and to exclude the possibility of task-specific effects, we conducted an associative mismatch task in a separate cohort of rats that examined a different type of behavioural flexibility, but also requiring inhibition. The associative mismatch task has been less comprehensively specified in terms of underlying neurobiology, with some evidence of involvement of hippocampal circuitry 42 , but it requires attentional, memory and inhibitory processes that have been shown to be mediated by callosal circuitry [51][52][53] . We again observed highly specific effects on inhibition of previous learning and no effects on learning per se. Together, the data from the two tasks provided converging evidence for an inhibitory deficit associated with Cyfip1 haploinsufficiency where prior learning dominates current responses leading to maladaptive inflexible behaviour. An inability to alter behaviour in response to changes in the environment has been strongly associated with ventral prefrontal . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; cortex damage in humans 54 and rats 55,56 . The corpus callosum (along with the internal capsule) carries white matter bundles containing axons projecting from the prefrontal cortex and striatal regions 19 , and it was shown that pruning and myelination of the corpus callosum coincides with cortical maturation in the prefrontal cortex, mutually influencing each other's development 57,58 .
Increases in FA with little evidence of reductions were a prominent feature of our findings in human 15q11.2 BP1-BP2 deletion carriers (in press) 7 . Hence, while the rat model and human phenotype converged on white matter changes they differed in the direction of the change.
Differences between the human and rat findings could have resulted from several factors.
First, the 15q11.2 BP1-BP2 deletion involves three other genes in addition to CYFIP1, especially NIPA1 which is expressed in the brain and was found to inhibit the bone morphogenic protein (BMP), crucial for typical axonal growth and guidance 59,60 . Therefore, a priori, haploinsufficiency of NIPA, and possibly the other genes in the interval 61,62 , may impact on the 15q11.2 BP1-BP2 deletion DTI phenotype. The possibility that there are species differences in the expression patterns of CYFIP1/Cyfip1 and also any compensatory responses to haploinsufficiency should also be borne in mind. Furthermore, the humans and rats are likely to have been subject to differential compensatory mechanisms arising from very different environmental challenges across their lifespan 6,63,64 . Moreover, as changes in myelin thickness impact relatively modestly on DTI measures 30 it may be that whilst myelin changes may be present in both human and the rat model, in terms of the human DTI data any effect on myelin may have been masked by other molecular and cellular consequences of the copy number deletion. To date, there have been no published studies of myelin (as opposed to overall white matter) changes in 15q11.2 BP1-BP2 deletion, though the current data predicts their existence and this is something that could be tested using ultrastructural magnetic . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; resonance imaging (MRI) methods providing the necessary resolution to visualise and quantify myelinated axons directly in the living human brain 65 . Nonetheless, whilst an exact between-species comparison would require an assessment of CYFIP1-specific heterozygous humans, we have demonstrated a clear link between Cyfip1 haploinsufficiency and white matter microstructure.
In conclusion, we have employed a novel rat model of Cyfip1 haploinsufficiency to probe the neurobiological and behavioural mechanisms underlying the significantly enhanced risk for psychopathology linked to the 15q11.2 BP1-BP2 deletion. We found disturbances to white matter as seen in human carriers, and showed effects on myelin thickness and a reduction in mature oligodendrocyte number, together with evidence of highly specific behavioural deficits due to maladaptive inhibition. The imaging phenotype on both rats and humans further suggest that it is unlikely that effects mediated by CYFIP1 are solely responsible, and additional work is required to determine the contribution made by the other three genes, NIPA1, NIPA2, and TUBGCP5 affected in the 15q11.2 BP1-BP2 deletion. However these findings on the Cyfip1 rat model give an insight into the contribution made by low dosage of CYFIP1 to the 15q11.2 BP1-BP2 deletion phenotype.
. CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

Competing financial interests
The authors declare no competing financial interests.

Material and correspondence
All data from this study are available from the corresponding author upon reasonable request.
. CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.  . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. Cardiff the rats were housed in mixed-genotype groups of 2-3 rats. The rats had free access to food and water (except for those used in the reversal learning task, see below) and lived under the condition of a 12 hr light/day cycle (lights on at 7:00 am), room temperature 21±2 °C. Rats used in DTI were 5 months old. The rats were euthanized 1 month after the scanning and used for immunofluorescence. The rats used for electron microscopy were 6 months old.
The rats used for behavioural experiments were 6-9 months old. The reversal learning task was motivated by liquid reward (10% sucrose solution w/v) and to enhance working in the task, the rats were subject to water restriction immediately prior to and during task training, in which case the rats were given 2 hours access to water per day. The water restriction schedule has no adverse effects on the health or welfare of the rats, being designed to give rise to a temporary increase in motivation for the liquid reinforcement, and across the whole day the rats on the schedule drink as much fluid as under free access conditions. All the experimental procedures were performed in accordance with institutional animal welfare and ARRIVE guidelines and the UK Home Office License PPL 30/3135. . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. minutes. In addition, high resolution, T2 weighted images were acquired for anatomical reference with a multi-slice multi-echo pulse sequence with the following parameters: TR of 7200ms, TE of 15ms and effective TE of 45ms, rare factor was 8. Image resolution was set to 0.22 mm 3 with matrix size of 128´160´50 to cover the entire brain.
DTI data correction and DTI maps extraction. ExploreDTI 4.8.3 66 and SPM (version 12, UCL, London, UK) were used in the preprocessing of the rat DTI data. First, eddy-current induced distortion and motion correction were performed, and mean-DWI images were extracted using ExploreDTI. Non-brain tissue was removed from the mean-DWI and the T2weighted images following these steps: (1) T2-weighted scans were anisotropic smoothed using ExploreDTI, (2) both smoothed T2-weighted and mean DWI images were bias corrected using the segmentation tool in SPM12, (3) the bias corrected T2-weighted were coregistered with a population-specific template and multiplied by a mask to remove the non-. CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; brain tissue, (4) the skull was removed from the mean DWIs using the 3D masking option in ExploreDTI. Then, data was corrected for field inhomogeneities, using ExploreDTI, where the skull-stripped mean DWIs images were used as a native space mask, and the skullstripped T2-weighted structural scans were used as transformed space mask. Each DWI image was nonlinearly warped to the T2-weighted image using non-DWIs map as a reference.
ExploreDTI was used to generate FA, AD, RD and MD maps.

Preprocessing for Tract-Based Spatial Statistics (TBSS).
For the voxel-wise analyses of DTI data, Tract-Based Spatial Statistics (TBSS) method was implemented, which is part of the FMRIB software library (FSL). All FA maps were submitted to a free-search for a best registration target, where each volume was first registered to every other volume, and the one requiring minimum transformation to be registered to other volumes was selected as the best registration target. This target was used as a template into which the registration was performed. Following registration, a mean FA map was calculated, thinned to represent a mean FA skeleton, and an optimal threshold of 0.2 was applied to the mean FA skeleton to create a binary white matter skeleton mask (Supplementary Figure 1). The local FA-maxima, as well as the AD, RD, and MD, of each rat were projected onto this white matter skeleton.
Transmission electron microscopy and immunofluorescence. For transmission electron microscopy, a new cohort of 9 rats (WT n=5 and Cyfip1 +/-n=4) was used. For immunofluorescence seven brains were randomly selected (WT n=7 and Cyfip1 +/-n=7) from the cohort used for DTI. In both cohorts, the rats were intracardially perfused with 0.1 M phosphate buffered saline (PBS), followed by 4% of glutatradehyde in 0.1 M PBS in the cohort used for electron microscopy, and 4% paraformaldehyde in 0.1 M PBS (PFA) in the cohort used for immunofluorescence. For transmission electron microscopy, the brains . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. were placed on a shaker to postfix in glutatradehyde for 4h, after which they were placed in phosphate buffered saline and stored at 4C until further use. Then the brains were embedded in TAAB embedding resin. Ultra-thin sections (50 nm) were stained with aqueous 4% uranyl acetate and lead citrate. The sections were visualized on a transmission electron microscope (CM12, Philips, the Netherlands) and, for quantification, images were taken using an on-axis 2048×2048 charge-coupled device camera (Proscan, Schering, Germany). In order to obtain a representative sample, 15 regions across the extent of the anterior-posterior extent of the corpus callosum per animal were taken for quantification. For immunofluorescence, the brains were placed on a shaker to postfix in PFA for 4 h, after which they were placed in phosphate buffered 30% sucrose. Coronal cryosections of the brain, of 15 µm thickness, were made on a cryostat (CM1860 UV, Leica, UK), mounted onto a polylysine coated slides (3 sections per slide), and stored at -20 ºC. For immunofluorescence, antibodies were used as follows: anti-Olig2 (ab109186, Abcam) 1:400, anti-APC [CC-1] (ab16794, Abcam) 1:400, anti-MBP (MAB386, Millipore) 1:300. For the Olig2 and Cc1 doublestaining, the slices were heated in a 5% citrate buffered antigen retrieval solution (pH 6, 10x, Sigma-Aldrich Company, UK), using a water bath at 90% for 10 min. All the slices were blocked for 1 hour with 5% donkey serum (Sigma-Aldrich Company, UK), and 0.3% Triton X-100 in PBS. The appropriate primary antibodies were applied and incubated overnight at 4 °C. On the next day, after washing, the slices were incubated for 2 hours with secondary antibodies (Alexa Fluor Life Technologies, Manchester, UK), in a concentration of 1:1000 at room temperature.
For quantification of Olig2+ and Cc1+ cells, images were taken on an inverted fluorescent time lapse microscope (DMI6000B, Leica, UK), and at least 4 images from random visual fields were taken from regions including the corpus callosum and external capsule. For . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. quantification of MBP intensity, one coronal section per rat was taken on an Axio scan (Zeiss, Germany), and the same exposure time and intensity were used for all the slides.
Reversal Learning. A separate group of 17 rats were used in the reversal learning task (WT n=7, Cyfip1 +/-n=10). Testing was conducted in a touchscreen-based automated operant system that consisted of an operant chamber with a flat-screen monitor equipped with an infrared touchscreen with accompanying Animal Behaviour Environment Test (ABET) II software (Campden Instruments, Leics). Session duration was 30 minute, or until 100 trials were completed under all training conditions. Pre-training consisted of two stages (Magazine Training and Touch Training) these gradually shaped the screen-touching behaviour required for the reversal learning touchscreen task proper. Following successful completion of pretraining Visual Discrimination Training began (Supplementary Figure 4); two stimuli were presented at a time (S+ and S-, counterbalanced across animals), on either side of the screen.
The rat had to touch the correct stimulus (S+) to elicit reward. Reward delivery was accompanied by illumination of the tray light and a tone. Entry to collect the reward turned off the tray light and started the inter-trial interval (ITI -5s) following which the rat initiated the next trial by a second magazine entry. Touching the incorrect stimulus (S-) terminated the trial and the house-light was turned on for a time-out period of 5s and no reward given, following the time-out the ITI period began after which the rat had to initiate the next trial by executing a magazine entry. Once the rats reached performance criteria (Completing 50+ trials with 80-85% correct, for 2 consecutive sessions), the contingencies were reversed (previous S+ now S-; previous S-now S+) and behaviour monitored (see Supplementary   Figure 4 for schematic of task design).
. CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. Mismatch Task. A different group of 36 rats performed the mismatch task (WT n=21, Cyfip1 +/-n=15). On the first 4 days, the rats were placed in the experimental apparatus (a modified skinner box allowing presentations of auditory and visual stimuli) for 30 min.
Following this general habituation to the apparatus they received 4 days of training with two audiovisual sequences. One auditory stimulus (a 2 kHz tone) preceded the constant presentation of a light, whereas a second auditory stimulus (a 10 Hz series of clicks) preceded the flashing presentation of the same light stimuli (i.e. ToneàSteady Light, ClickàFlashing Light; the combinations were randomly counterbalanced across animals). All stimuli were presented for 10 sec. There were 10 presentations of both audiovisual sequences on each of the first 3 days of training and six presentations of both sequences on day 4 that served as warm-up trials for the eight test trials that immediately followed. The inter-trial-interval was 2 min. Rats received two types of test trials, match and mismatch. The order in which the two types of test trials were presented was counterbalanced. Match test trials were presentations of the same audiovisual sequences that had been presented during training (e.g. ToneàSteady Light, ClickàFlashing Light), whereas on mismatch trials the auditory stimuli preceding the visual stimuli were exchanged (e.g. ToneàFlashing Light, ClickàSteady Light). All experimental sessions were recorded using a video recorder and orienting responses subsequently scored by observers who were blind to the genotype of the rats and the nature of the test trials (match or mismatch). An Orienting Response (OR) was defined as the tip of a rat's snout being located in the side of the apparatus that contained the light and pointing in the direction of the light.
Statistical analyses. Differences in DTI measures between the two groups (WT and Cyfip1 +/-) were assessed using voxel-wise independent t-tests, where two different contrasts were used (WT>Cyfip1 +/-, and Cyfip1 +/->WT). Using the randomize function (part of FSL), the null . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; distribution was built over 1000 random permutations, using the Threshold-Free Cluster Enhancement (TFCE) 25 algorithm where cluster-like structures are enhanced, and the results are shown for p<0.05. For multiple comparison correction, first FWE correction was used.
Since only FA changes were found within this analysis, we also used a less conservative correction method based on FDR correction, purposed by Benjamini-Hochberg 26 . To quantify the changes in areas where significant differences in FA were seen after FWE correction, regions of interest (ROIs) were manually delineated using FSL. Several consecutive slices were outlined on the coronal plane and the selected ROIs included the corpus callosum, internal capsule, external capsule, and fornix/fimbria regions. The CBJ13 MR-histology rat atlas at age P80 68 was used as reference. A representation of the binary masks can be found in Supplementary Figure 2. FA, AD, RD, and MD were quantified by applying these binary masks and extracting the mean values for each region across subjects.
For quantification of cells the ImageJ software (version 1.51) was used. The number of myelinated and unmyelinated axons, axon diameter, myelin thickness and g-ratio (measure of myelin thickness relative to axon diameter: where lower g-ratios indicates thicker myelin sheath) of normally myelinated axons were quantified. A total of 13128 (WT n=7148, Cyfip1 +/-n=5980 axons) myelinated axons were analysed. For quantification of oligodendrocytes, the total number of Olig2+, and the overlapped Olig2+/Cc1+ cells were counted. Only cell bodies clearly identified by Olig2 and Cc1 immunofluorescence and overlapping with DAPI staining were counted. The number of cells were divided by the area quantified in each image. MBP+ reactivity was determined by comparing immunofluorescence staining intensity. The whole region of corpus callosum and external capsule was selected in the coronal section, and quantification was done by calculating the mean intensity of the pixels above a preset intensity threshold, multiplied by the number of . CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/477786 doi: bioRxiv preprint first posted online Nov. 25, 2018; pixels above that threshold, and dived by the total area quantified. All the analyses were conducted with the investigator blinded to the phenotype. Differences between WT and Cyfip1 +/were analysed in RStudio. In order to compare all the axons in each group while taking into account variation across individuals, we used linear mixed effects models to analyse the effect of genotype on axon diameter, g-ratio and myelin thickness, where these measures were considered fixed effects, and animals were considered random effects. Since we only had one random effect, we used non-restricted maximum likelihood to estimate the model parameters. In this analysis, the myelin thickness was log-transformed since the data followed a log-normal distribution, whereas the other measures followed a normal distribution. All the other measures were analysed using two-tailed unpaired Student's t-test.
Data are given as mean ± s.e.m.
Visual discrimination and reversal learning performance was assessed using ANOVA with factors of GENOTYPE and SESSION. Any significant interaction was subsequently examined by analysing the Simple Effects. Completion rates for the rats during the different phases of reversal were assessed non-parametrically using Chi-squared test. Performance in the mismatch task was assessed using ANOVA with factor of GENOTYPE and BLOCK during the habituation to the test apparatus phase of training and factor GENOTYPE in the habituation to the stimulus pairs and mismatch test phases. Orienting responses in the mismatch test phase were analysed as a discrimination ratio (total orienting to matched / total orienting to both matched + mismatched).
. CC-BY 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.